Spelling suggestions: "subject:"bayesian metaparameter destimation"" "subject:"bayesian metaparameter coestimation""
1 |
Mathematical and Statistical Insights in Evaluating State Dependent Effectiveness of HIV Prevention InterventionsJanuary 2014 (has links)
abstract: Pre-Exposure Prophylaxis (PrEP) is any medical or public health procedure used before exposure to the disease causing agent, its purpose is to prevent, rather than treat or cure a disease. Most commonly, PrEP refers to an experimental HIV-prevention strategy that would use antiretrovirals to protect HIV-negative people from HIV infection. A deterministic mathematical model of HIV transmission is developed to evaluate the public-health impact of oral PrEP interventions, and to compare PrEP effectiveness with respect to different evaluation methods. The effects of demographic, behavioral, and epidemic parameters on the PrEP impact are studied in a multivariate sensitivity analysis. Most of the published models on HIV intervention impact assume that the number of individuals joining the sexually active population per year is constant or proportional to the total population. In the second part of this study, three models are presented and analyzed to study the PrEP intervention, with constant, linear, and logistic recruitment rates. How different demographic assumptions can affect the evaluation of PrEP is studied. When provided with data, often least square fitting or similar approaches can be used to determine a single set of approximated parameter values that make the model fit the data best. However, least square fitting only provides point estimates and does not provide information on how strongly the data supports these particular estimates. Therefore, in the third part of this study, Bayesian parameter estimation is applied on fitting ODE model to the related HIV data. Starting with a set of prior distributions for the parameters as initial guess, Bayes' formula can be applied to obtain a set of posterior distributions for the parameters which makes the model fit the observed data best. Evaluating the posterior distribution often requires the integration of high-dimensional functions, which is usually difficult to calculate numerically. Therefore, the Markov chain Monte Carlo (MCMC) method is used to approximate the posterior distribution. / Dissertation/Thesis / Doctoral Dissertation Applied Mathematics 2014
|
2 |
Minimally Corrective, Approximately Recovering Priors to Correct Expert Judgement in Bayesian Parameter EstimationMay, Thomas Joseph 23 July 2015 (has links)
Bayesian parameter estimation is a popular method to address inverse problems. However, since prior distributions are chosen based on expert judgement, the method can inherently introduce bias into the understanding of the parameters. This can be especially relevant in the case of distributed parameters where it is difficult to check for error. To minimize this bias, we develop the idea of a minimally corrective, approximately recovering prior (MCAR prior) that generates a guide for the prior and corrects the expert supplied prior according to that guide. We demonstrate this approach for the 1D elliptic equation or the elliptic partial differential equation and observe how this method works in cases with significant and without any expert bias. In the case of significant expert bias, the method substantially reduces the bias and, in the case with no expert bias, the method only introduces minor errors. The cost of introducing these small errors for good judgement is worth the benefit of correcting major errors in bad judgement. This is particularly true when the prior is only determined using a heuristic or an assumed distribution. / Master of Science
|
3 |
Inversion d’un modèle de culture pour estimer spatialement les propriétés des sols et améliorer la prédiction de variables agro-environnementales / Inversion of a crop model for estimating spatially the soil properties and improving the prediction of agro-environmental variablesVarella, Hubert Vincent 15 December 2009 (has links)
Les modèles de culture constituent des outils indispensables pour comprendre l’influence des conditions agropédoclimatiques sur le système sol-plante à différentes échelles spatiales et temporelles. A l’échelle locale de la parcelle agricole, le modèle peut être utilisé dans le cadre de l’agriculture de précision pour optimiser les pratiques de fertilisation azotée de façon à maximiser le rendement ou le revenu tout en minimisant le lessivage des nitrates vers la nappe. Cependant, la pertinence de l’utilisation du modèle repose sur la qualité des prédictions réalisées, basée entre autres sur une bonne détermination des paramètres d’entrée du modèle. Dans le cadre de l’agriculture de précision, les paramètres concernant les propriétés des sols sont les plus délicates à connaître en tout point de la parcelle et il existe très peu de cartes de sols permettant de les déterminer de manière précise. Néanmoins, dans ce contexte, on peut disposer d’observations acquises automatiquement sur l’état du système sol-plante, telles que des images de télédétection, les cartes de rendement ou les mesures de résistivité électrique du sol. Il existe alors une alternative intéressante pour estimer les propriétés des sols à l’échelle de la parcelle qui consiste à inverser le modèle de culture à partir de ces observations pour retrouver les valeurs des propriétés des sols. L’objectif de cette thèse consiste (i) dans un premier temps à analyser les performances d’estimation des propriétés des sols par inversion du modèle STICS à partir de différents jeux d’observations sur des cultures de blé et de betterave sucrière, en mettant en oeuvre une méthode bayésienne de type Importance Sampling, (ii) dans un second temps à mesurer l’amélioration des prédictions de variables agro-environnementales réalisées par le modèle à partir des valeurs estimées des paramètres. Nous montrons que l’analyse de sensibilité globale permet de quantifier la quantité d’information contenue dans les jeux d’observations et les performances réalisées en matière d’estimation des paramètres. Ce sont les propriétés liées au fonctionnement hydrique du sol (humidité à la capacité au champ, profondeur de sol, conditions initiales) qui bénéficient globalement de la meilleure performance d’estimation par inversion. La performance d’estimation, évaluée par comparaison avec l’estimation fournie par l’information a priori, dépend fortement du jeu d’observation et est significativement améliorée lorsque les observations sont faites sur une culture de betterave, les conditions climatiques sont sèches ou la profondeur de sol est faible. Les prédictions agro-environnementales, notamment la quantité et la qualité du rendement, peuvent être grandement améliorées lorsque les propriétés du sol sont estimées par inversion, car les variables prédites par le modèle sont également sensibles aux propriétés liées à l’état hydrique du sol. Pour finir, nous montrons dans un travail exploratoire que la prise en compte d’une information sur la structure spatiale des propriétés du sol fournie par les mesures de résistivité électrique, peut permettre d’améliorer l’estimation spatialisée des propriétés du sol. Les observations acquises automatiquement sur le couvert végétal et la résistivité électrique du sol se révèlent être pertinentes pour estimer les propriétés du sol par inversion du modèle et améliorer les prédictions des variables agro-environnementales sur lesquelles reposent les règles de choix des pratiques agricoles / Dynamic crop models are very useful to predict the behavior of crops in their environment and are widely used in a lot of agro-environmental work. These models have many parameters and their spatial application require a good knowledge of these parameters,especially of the soil parameters. These parameters can be estimated from soil analysis at different points but this is very costly and requires a lot of experimental work. Nevertheless,observations on crops provided by new techniques like remote sensing or yield monitoring, is a possibility for estimating soil parameters through the inversion of crop models. In my work, the STICS crop model is studied for the wheat and the sugar beet and it includes more than 200 parameters. After a previous work based on a large experimental database for calibrate parameters related to the characteristics of the crop, I started my study with a global sensitivity analysis of the observed variables (leaf area index LAI and absorbed nitrogen QN provided by remote sensing data, and yield at harvest provided by yield monitoring) to the soil parameters, in order to determine which of them have to be estimated. This study was made in different climatic and agronomic conditions and it reveals that 7 soil parameters (4 related to the water and 3 related to the nitrogen) have a clearly influence on the variance of the observed variables and have to be therefore estimated. For estimating these 7 soil parameters, I chose a Bayesian data assimilation method (because I have prior information on these parameters) named Importance Sampling by using observations, on wheat and sugar beet crop, of LAI and QN at various dates and yield at harvest acquired on different climatic and agronomic conditions. The quality of parameter estimation is then determined by comparing the result of parameter estimation with only prio rinformation and the result with the posterior information provided by the Bayesian data assimilation method. The result of the parameter estimation show that the whole set of parameter has a better quality of estimation when observations on sugar beet are assimilated. At the same time, global sensitivity analysis of the observed variables to the 7 soil parameters have been performed, allowing me to build a criterion based on sensitivity indices (provided by the global sensitivity analysis) able to rank the parameters with respect to their quality of estimate. This criterion constitutes an interesting tool for determining which parameters it is possible to estimate to reduce probably the uncertainties on the predictions. The prediction of the crop behaviour when estimating the soil parameters is then studied. Indeed, the quality of prediction of agro-environmental variables of the STICS crop model (yield, protein of the grain and nitrogen balance at harvest) is determined by comparing the result of the prediction using the prior information on the parameters and the result using the posterior information. As for the estimation of soil parameters, the prediction of the variable is made on different climatic and agronomic conditions. According to the result of parameter estimation, assimilating observations on sugar beet lead to a better quality ofprediction of the variables than observations on wheat. It was also shown that the number ofcrop seasons observed and the number of observations improve the quality of the prediction
|
Page generated in 0.1297 seconds